Millions of Americans are receiving behavioral interventions for problematic alcohol use. In 2010, the Substance Abuse and Mental Health Services Administration (SAMHSA) documented over 1.8 million treatment episodes for drug and alcohol problems, many involving group or individual psychotherapy. The tremendous service-delivery need has focused research on optimal training methods, to promote the dissemination of evidence-based interventions. A recent meta-analysis of motivational interviewing (MI) shows that post-training supports - such as performance-based feedback or coaching - are critical for maintaining counselor skills following training. However, the practical implementation of performance-based feedback for alcohol use disorders (AUDs) and problematic drinking is currently prohibitive in effort, time, and money. There is a critical need or technology to scale up performance-based feedback to counselors for AUDs and problematic drinking. This competitive renewal builds on interdisciplinary research focused on automating the evaluation of MI fidelity for alcohol and substance use problems. This collaborative research brings together speech signal processing experts from electrical engineering and statistical text-mining and natural language processing experts from computer science with MI expert trainers and researchers. Our previous research laid a computational foundation for generating MI fidelity codes from semantic and vocal features, and the current proposal moves this work into direct clinical application. In collaboration with the University of Utah Counseling Center (UCC), we will develop and implement a clinical software support tool, the Counselor Observer Ratings Expert for MI (CORE-MI). The CORE-MI system will provide performance-based feedback focused on MI fidelity codes for training, supervision, and quality assurance for counselors treating clients struggling with alcohol and substance use problems. The research will use a hybrid implementation-effectiveness design to pursue the following three aims: 1) Implement and calibrate the CORE-MI system at the UCC clinic to provide automated, performance-based feedback on MI; 2) Compare counselor fidelity to MI and client alcohol and substance use outcomes, before and after initiation of the CORE-MI system (approximately, N = 2,400 sessions); and 3) Using machine learning tools, computationally explore mechanisms of MI using semantic and vocal data, MI fidelity codes, and client outcomes from approximately 3,000 sessions. The successful execution of this project will break the reliance on human judgment for providing performance-based feedback to MI and will massively expand the capacity to train, supervise, and provide quality assurance.